Research map / Image generation / Fast diffusion sampling
Di[M]O: Distilling Masked Diffusion Models into One-step Generator
Yuanzhi Zhu, Xi Wang, Stéphane Lathuilière, Vicky Kalogeiton
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| distill the MaskGit teacher into a one-step generator with Di[M]O | MaskGit · ImageNet-256 | FID: 16-step teacher 6.60; Di[M]O 1 step 6.91 | Improved |
| one-step Di[M]O instead of 4-step teacher sampling | MaskGit · ImageNet-256 | FID: 10.73 -> 6.91 | Improved |
| distill Meissonic into one-step generator with Di[M]O | Meissonic · HPSv2 averaged | HPS score: 4-step teacher 24.66 -> 28.11 | Improved |
| Di[M]O one step versus teacher with 48 steps | Meissonic · GenEval overall | overall score: 0.54 -> 0.43 | Hurt |
| Di[M]O one step versus other one-step diffusion generators | Di[M]O · HPSv2 averaged | HPS score: 27.48 -> 28.11 | Improved |
| set initial mask ratio r_init = 1 (all masked tokens) (ablation) | Di[M]O · ImageNet-256 | FID / training stability | Hurt |
| set initial mask ratio near 0 (all random tokens) (ablation) | Di[M]O · ImageNet-256 | FID / training stability | Hurt |
| use hybrid initialization with r_init = 0.6 (ablation) | Di[M]O · ImageNet-256 | FID | Improved |
| reduce Jeffrey coefficient beta, down to -0.2 (ablation) | Di[M]O · ImageNet-256 | FID | Improved |
| add Gaussian perturbation on initial token embeddings (ablation) | Di[M]O · ImageNet-256 | FID | Improved |
Closest papers by meaning
- Simplified and Generalized Masked Diffusion for Discrete Data (NeurIPS 2024 · similarity 0.92)
- Beyond Masked and Unmasked: Discrete Diffusion Models via Partial Masking (NeurIPS 2025 · similarity 0.92)
- Random Conditioning for Diffusion Model Compression with Distillation (CVPR 2025 · similarity 0.91)
- One-Step Diffusion Distillation through Score Implicit Matching (NeurIPS 2024 · similarity 0.91)
- Masked Diffusion Models are Secretly Time-Agnostic Masked Models and Exploit Inaccurate Categorical Sampling (ICLR 2025 · similarity 0.91)
- Simple and Effective Masked Diffusion Language Models (NeurIPS 2024 · similarity 0.91)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
